He's back! I'm sorry to hear about everything that happened during there months, I hope everything is better now and will improve, for both you and Dingus!
Thank you for the well wishes! I'm very time poor at the moment but I have a roof over my head and that means a lot! I can't keep to any sort of content schedule at the moment but when I have some time and ideas I will keep trying to make some more vids!
I really like this idea. Would be interesting if there was some sort of noise/hearing mechanic based on Dingus's speed and a proximity around the guards.
Hope ur doing alright i absolutely love your videos and seeing how terribly dingus does just know, no matter what me and whatever other people are here to support no matter what happens
I feel like Dingus kinda just took the exact strategy from the first level and opened with that in levels 2 and 3, (get to the edge of the map then rotate around the edge clockwise), and because the level geometry allowed for that he had really good looking starts but then with the third level he largely had to figure out the concept of “targets not on the edge of the map”
Unfortunately in a static environment that is very likely to happen, the AI is optimising for score and there is an optimal way to play the level, so he's "learning the level" not "learning to play the game" Injecting randomness into the game would certainly help address that, as would randomly swapping between levels. Actually I've been thinking about this after seeing a lot of comments on learning vs memorisation, I don't know if I'd make a full edited video but perhaps could do a livestream of some of the static level challenges randomly swapping between levels, which would force the AI to become more generalised. I would need to do a lot of testing with hyperparameters beforehand to make sure it continually makes progress, but it could be really cool.
I just want to say thank you for making these videos even though you don’t have to and it being hard for you to find the time to make these videos makes me respect you thank you ❤
Hmmm i think one of the problems is dingus has to "scan" the room to "see" all the variables coming his way -- what if you seperate his "eyesite" from his motions ?? Im probably wrong but it seems he has to bounce left and right to see the "relationships" in front of him
This is definitely my favorite vid from you, I can tell there was work put into the environment and models. Only really small thing I would recommend is using solid colors on the ground/walls, either blank or with and outline.
Little disappointed this isn't a real stealth game. The opposing agents aren't really intelligent and basically are just big orange fan shapes that move with a block affixed to their ends. And Dingus doesn't really have to plan or reason, because once he gets all the technology he just wins even if he's frames from getting captured. I'd like to see what would happen if he had a safe starting position that'd punish him for idling in it, a time limit, a score for getting the tech, and then a extraction point to escape to. The guards could be fixed or on a path with randomized starting positions forcing Dings not merely to go through the motions of a plan but to react to his environment. Maybe you could use his position on the map as input data or something? This sounds a little beyond the scope of this research but the changing variables might make it more interesting. Especially if there were two agents competing against one another or forced to collaborate. It'd make it kind of interesting to see how behaviors develop naturally in response to these adversarial and potentially collaborative conditions.
I would love to see those agents training as well. What would happen? Would they adapt indifinitely? Would one find the ultimate advantage over the other? And it is easy to make a score function: agents score is just 1 - dingus (assuming Dingus' score to be 0-1).
You should give these puzzles to a human, but give them less perception than Dingus gets. Maybe just a straight line in front of the player that tells if it's empty, enemy vision, wall, or tech?
Okay, but what are these rewards? What is a positive reward for a lil ai guy? Whats a negative reward for the lil guy? Is it just telling him good or bad?
i love how when he runs out of time the game says “too dang slow dingus”
Next try to train the agents to catch Dingus.
I have played around with a few multi agent scenarios, they're tougher to test but I have a few potentially fun ideas
And then back and forth and back and forth until it is maximum level
Or at the same time in the same rooms as eachother
ok
@@Dingus_Labsteach dingus to swat team
Otacon: "Snake, AI is taking over the espionage industry! We need to unionize!"
Snake: "Haha, funny robot."
He's back!
I'm sorry to hear about everything that happened during there months, I hope everything is better now and will improve, for both you and Dingus!
Thank you for the well wishes!
I'm very time poor at the moment but I have a roof over my head and that means a lot!
I can't keep to any sort of content schedule at the moment but when I have some time and ideas I will keep trying to make some more vids!
@@Dingus_LabsI love your content man!
The little dingus that could ❤
Barely could.
0:38 As a nuclear engineering student, I can confidently say that mailing fission is not something that you want to do.
WE GETTING OUT OF THE MILITARY WITH THIS ONE🗣🗣🗣🔥🔥🔥💯💯💯
This is what Snake's VR Training was like.
Next Episode: Teaching Dingus to Achieve World Domination
I really like this idea. Would be interesting if there was some sort of noise/hearing mechanic based on Dingus's speed and a proximity around the guards.
Sorry to hear about the ordeal you're going through.
i cant believe dingus broke into your house irl
ai is getting out of hand!
Hope ur doing alright i absolutely love your videos and seeing how terribly dingus does just know, no matter what me and whatever other people are here to support no matter what happens
I think the guards are in love they all follow each other.
Wow! Great video. This channel is a hidden gem.
you should add an attempt counter
Fission mailed - dingus labs 2023
I feel like Dingus kinda just took the exact strategy from the first level and opened with that in levels 2 and 3, (get to the edge of the map then rotate around the edge clockwise), and because the level geometry allowed for that he had really good looking starts but then with the third level he largely had to figure out the concept of “targets not on the edge of the map”
Dingus is gonna end up being so smart one day from these
I feel like it didn't really "learn" much rather than find optimal routes.
If any randomness came into play, I think it would entirely break this AI.
Unfortunately in a static environment that is very likely to happen, the AI is optimising for score and there is an optimal way to play the level, so he's "learning the level" not "learning to play the game"
Injecting randomness into the game would certainly help address that, as would randomly swapping between levels.
Actually I've been thinking about this after seeing a lot of comments on learning vs memorisation, I don't know if I'd make a full edited video but perhaps could do a livestream of some of the static level challenges randomly swapping between levels, which would force the AI to become more generalised.
I would need to do a lot of testing with hyperparameters beforehand to make sure it continually makes progress, but it could be really cool.
Well I mean finding optimal routes IS learning. Learning how to maneuver creatively through obstacles and efficiently do it
I just want to say thank you for making these videos even though you don’t have to and it being hard for you to find the time to make these videos makes me respect you thank you ❤
Hmmm i think one of the problems is dingus has to "scan" the room to "see" all the variables coming his way -- what if you seperate his "eyesite" from his motions ??
Im probably wrong but it seems he has to bounce left and right to see the "relationships" in front of him
This is definitely my favorite vid from you, I can tell there was work put into the environment and models. Only really small thing I would recommend is using solid colors on the ground/walls, either blank or with and outline.
You should give dingus a brother with a similar name
Old mate Dangus
DANGUS, YES. @@Dingus_Labs
Cube ai’s seem to be getting quite popular, I like it!
This is amazing thank you for the content this made my day!
Little disappointed this isn't a real stealth game. The opposing agents aren't really intelligent and basically are just big orange fan shapes that move with a block affixed to their ends.
And Dingus doesn't really have to plan or reason, because once he gets all the technology he just wins even if he's frames from getting captured.
I'd like to see what would happen if he had a safe starting position that'd punish him for idling in it, a time limit, a score for getting the tech, and then a extraction point to escape to.
The guards could be fixed or on a path with randomized starting positions forcing Dings not merely to go through the motions of a plan but to react to his environment.
Maybe you could use his position on the map as input data or something?
This sounds a little beyond the scope of this research but the changing variables might make it more interesting.
Especially if there were two agents competing against one another or forced to collaborate. It'd make it kind of interesting to see how behaviors develop naturally in response to these adversarial and potentially collaborative conditions.
I would love to see those agents training as well. What would happen? Would they adapt indifinitely? Would one find the ultimate advantage over the other? And it is easy to make a score function: agents score is just 1 - dingus (assuming Dingus' score to be 0-1).
you should compile all of these AIs to make the prime DINGUS.
You should give these puzzles to a human, but give them less perception than Dingus gets. Maybe just a straight line in front of the player that tells if it's empty, enemy vision, wall, or tech?
Hope ur feelin alright now m8. Keep up the great work
Just found this channel, glad you got your standing back after all that stuff that happened. Like the content though, keep it up!
One Dingus, Many Dingi
AI learns to Metal Gear Solid
I love your videos so much! Do you ever share the code/projects that you make?
Geometry dash. 3:14
Hey you should teach dingus that he also has to escape the level through an exit as well
Now combine the agent one with the stealing one
or will this be a case of "Fisson Mailed" lol
ans then, after hundreds of attempts, dingus got this run
God speed mr.dingus
I love dingus, he's cute!
I hope dingus won't get the key to the real world one day 🙏🏻
What is the song at 1:30?
@Dingus_Labs do you publish the projects in your github?
Vision Mailed _👍_
I hope things only go up from here for you!
What a little Dingus, amirite? :p
I need a dingus plushie
Do you think you could give dingus legs and arms and teach him how to dance?
Dingus who sold the world
mmm, comment to boost this man's algorithm
What if you strapped Dingus into a crappy car and let Dingus drive the boat
How come there aren’t any preburglary videos?
When he said "fission mailed", I'm pretty sure it's a reference to Henry stickman 0:37
it's a reference to metal gear solid, which henry stickmin also referenced
Dingus Dond
❤ great video
Good job!
first this... than an AI to steal things...
anything you want to tell us?
how do you do this could you make a tutorial
Dingus
Great video
how are you not famous lmao
I don't have a clue what I'm doing! I'm glad some peeps are enjoying it!
Do you think you could, say, leak the code for the final product?
No real reason for my asking…
yipee
Cool
Åh hej dingus
Okay, but what are these rewards? What is a positive reward for a lil ai guy? Whats a negative reward for the lil guy? Is it just telling him good or bad?
You just put an ai in the vr simulations
Great video